AI prices the risk on every trade
The model computes this counterparty’s probability of default on this trade and sets the collateral from it.
- Post a fraction, pay the balance on delivery
- The more you settle, the less you post
Atara is a conditional payment protocol. It computes the risk on every payment and takes a fraction up front instead of the whole sum.
Pricing needs somebody to decide the condition was met, and both of them need a record of what actually happened.
The model computes this counterparty’s probability of default on this trade and sets the collateral from it.
Whether the condition was met is decided on what the evidence says.
Every completed trade leaves one record with a real amount attached: who paid on time, who defaulted.
A payment moves down the stack, and every layer writes into the record at the bottom. Adjudication is the only layer that can send a settlement back up.
One surface to integrate against.
The normal path. Many readers, one signed verdict.
Funds held by contract, not by us.
The last word when a condition is disputed — and the only ruling that can roll a settlement back.
Every settlement leaves a row.
Four layers route, judge, execute and record one trade.
Each answers to evidence, not to us.
One surface takes the order, the limits and the reconciliation, and decides where the trade settles. It is the only thing you integrate against.
One API key, one order. You never pick a chain or a venue.
Every trade is read twice — one model prices the counterparty's risk of default, the other rules on whether the condition was met. Neither decides alone: a verdict takes M-of-N signatures and carries the trace of the evidence it was built from.
No single model — and no single operator — can decide your case.
Funds sit in an escrow contract with three states — hold, release, rollback — and a deadline that block time enforces, not us. The contract is deployed on the chain the counterparty already settles on, so value never leaves the chain it started on.
With every Atara service down, both sides can still withdraw. No bridge.
Every settlement writes one row — who paid, who disputed, how it ruled — written by the settlement itself, not by either side. Each row is linked to the events that caused it, and the credit model reads those rows on the next trade.
The second trade is priced on the first, from settlements rather than self-reported scores.
Settlement history, dispute outcomes, verification checks, and where the funds came from.
For this counterparty, on this trade. Not a tier set when the account opened.
What you post, what you pay, and on what evidence it unwinds.
A limit set when the account opens says nothing about who is on the other side of this one.
Other systems answer whether you may proceed. This answers what you post and what you pay.
How a trade unwinds is fixed before either side commits, not negotiated after something breaks.

A clause like “pay on acceptable quality” is broken down, before the trade, into standards both sides accept, each bound to a named source of evidence: a document, a reading, or a test result. The ruling is about what the source says, not about what either side believes.
Today, each side reads “acceptable” its own way once money is at stake, and the reading comes off one party’s own instrument.

Whichever way a case is decided, the judge earns the same. Forfeited collateral never flows to the judge.
Today, a dispute goes to the custodian or the platform: the party that holds the money and takes a cut of the trade it is asked to rule on.

Objective evidence such as a bank receipt triggers the release. A counterparty that owes performance cannot stall your money by withholding confirmation.
Today, you deliver, the other side declines to confirm, and the funds sit in escrow.
A size, a side, and a counterparty you do not trust yet.
StartThe Atara console is in private preview. Enter your code to open it.
No code? Write to hello@loka.cash.